HomebaseWorks
Lesson 1 of 4

Building a clear prompt: role, context, constraints, examples and format

A five-part structure for prompts that produce usable first drafts, with before and after examples from everyday work tasks.

EWritten by the HomebaseWorks editorial team · reviewed 26 Oct 2026

An AI chat assistant only knows what you tell it. Vague prompts produce generic answers, and you spend more time fixing them than you saved. A simple structure makes results much more predictable.

Advertisement

The five parts of a clear prompt

  1. Role: who the assistant should act as. "You are a support agent for an online furniture shop."
  2. Context: the facts it needs. The customer's issue, your policy, the audience.
  3. Constraints: limits such as length, tone, reading level, or things to avoid.
  4. Examples: one or two samples of the style you want. This is called few-shot prompting.
  5. Output format: a list, a table with named columns, an email with a subject line.

You will not need all five every time, but missing context and missing format cause most poor results.

Before and after

Weak: Write a reply to a customer about a late order.

Strong:

You are a support agent for an online furniture shop.
Customer: Ravi. His sofa order is 5 days late because of
a supplier delay. New delivery date: 14 March.
Write a reply under 120 words. Apologise once, give the
new date, offer free delivery on his next order.
Tone: friendly and plain. No exclamation marks.

The second version leaves almost nothing to guess, so the first draft is usually close to usable.

Separating instructions from material

When you paste a long text to summarise or rewrite, mark where it starts and ends, for example with a label like EMAIL: and quotation marks, and put your instruction before or after it. This stops the assistant from treating lines in the pasted text as instructions.

Iterating instead of restarting

If the answer is not right, say exactly what to change: "Shorten to 80 words", "Use bullet points", "Remove the second paragraph". Specific feedback works better than repeating the same prompt. Once a prompt works well for a recurring task, save it as a template so colleagues get consistent results.

Quick check before you send a prompt: does it say who the answer is for, what facts to use, how long it should be, and what shape it should come back in?

Advertisement
Continue with AI tools and prompting4 lessons · 1 interview set · 1 test · 1 certificate
Not tested

More in Data and AI